5. OpenClaw TUI Interaction

The TUI (Text User Interface) serves as OpenClaw's "command-line chat window." When it is inconvenient to access the web console on your Jetson device—or if you are connected remotely via SSH—the TUI allows you to perform most daily interactions directly within your terminal: engaging in dialogue, switching models, checking status, managing sessions, invoking skills, and more.

1. How to Access the TUI

Execute the following command in your terminal:

openclaw tui

If you have already launched the OpenClaw gateway (e.g., the gateway/server component) in a separate terminal session, you can simply run openclaw tui to enter the interactive interface.

For non-Docker versions, the OpenClaw gateway does not require manual startup; it launches automatically upon system boot.

For Docker versions (such as on the Jetson Nano), you must first open a terminal and manually run openclaw gateway to start the gateway service.

2. What is the TUI Interface?

You can think of the TUI as a "chat box with built-in commands":

• Upper Section: Dialogue History (messages you sent, model responses, and tool/skill outputs).
• Lower Section: Input Field (where you type questions or send slash commands).
• Sending Messages: Press Enter after typing to send your input.

When accessing the TUI for the first time, it is recommended that you perform two initial actions:

xxxxxxxxxx
/status

3. Slash Commands

Slash commands typically need to be sent as standalone messages beginning with a forward slash (/); they are used to control the current session and execute specific operations.

Two additional tips for a smoother "daily workflow":

• It is safest to include only one slash command per message.
• To quickly browse through all available commands, use /help or /commands.

4. Sessions and Memory

In OpenClaw, the roles of these two concepts are clearly defined:

• Session: Manages "where the current conversation stands." When you use /new or /reset within the TUI, you are essentially switching to—or restarting—a session.
• Memory: Manages "content that should be retained long-term, even across different sessions." This data is written to disk (in Markdown format) and can be retrieved back into the current session whenever needed. ### 4.1 Where Are Session Files Located?

Sessions are persisted to disk in JSONL format. Common locations include (subject to your actual installation path):

• Index: ~/.openclaw/agents/<agentId>/sessions/sessions.json
• Transcripts: ~/.openclaw/agents/<agentId>/sessions/<SessionId>.jsonl

For troubleshooting purposes, a more convenient approach is to first execute:

xxxxxxxxxx
openclaw status

This command typically displays the storage path as well as information regarding recent sessions.

4.2 How to "Put Memory to Use"

You can think of Memory as "having the bot take notes":

• memory/YYYY-MM-DD.md: A daily process log, suitable for recording progress and ad-hoc decisions.
• MEMORY.md: Long-term, stable information (preferences, fixed rules, project constraints, etc.).

Practical Tips:

• To make a specific conclusion accessible across multiple sessions: Simply tell the bot, "Write this into MEMORY.md."
• When preparing to switch to a new topic using /new: First use /compact—or write key decisions into Memory—before switching sessions.

5. Quick Start Workflow

1. Enter the TUI: openclaw tui
2. Check Status: /status
3. Select a Model: /models list → /model <Number or Name>
4. Daily Conversation: Simply type your question and press Enter.
5. Switch Topics: /new
6. Context Too Long: /compact
7. Preserve Key Conclusions Long-Term: Ask the bot to write them into Memory.

6. Troubleshooting Common Issues

6.1 Commands Are Unresponsive

• Ensure the command is sent on a separate line (do not mix it with regular text).
• Use /help first to check which commands are supported in your current version.

6.2 The Bot Starts "Rambling" or "Misunderstanding You," or Seems Distracted by History

• Start a new session: /new or /reset
• Check what context is currently being injected: /context detail

6.3 Responses Are Becoming Increasingly Slow

• Reduce "thinking depth": /think low or /think minimal
• Compress conversation history: /compact